Human activity recognition is gaining more and more the attention of researchers due to its applicability in many different fields such as health monitoring, smart environments, etc. Activity recognition solutions typically focus on the classification of single-user behavior. However, in a living or working environment, there are usually multiple inhabitants acting together, hence it makes sense to interpret the activities by considering the aggregated information from different subjects. In this paper, we address the problem of group activity recognition (GAR) in a hierarchical way by first examining individual person’s actions, reconstructed by correlating data coming from body-worn and external positioning sensors. We then aggregate this information by considering each individual as an input of a hierarchical deep belief network (DBN). This aims to extract common temporal/spatial dynamics at the level of group activity. We evaluated the proposed approach in a laboratory environment, where the participants labeled their daily activities using an app on a mobile phone. Collected data contributed to the creation of two datasets respectively containing labeled single and group activities. The experimental results evaluated on these datasets and on a public one demonstrated the effectiveness of the proposed model with respect to a support vector machine (SVM) baseline.

Working together: a DBN approach for individual and group activity recognition

Staffa M.
2020

Abstract

Human activity recognition is gaining more and more the attention of researchers due to its applicability in many different fields such as health monitoring, smart environments, etc. Activity recognition solutions typically focus on the classification of single-user behavior. However, in a living or working environment, there are usually multiple inhabitants acting together, hence it makes sense to interpret the activities by considering the aggregated information from different subjects. In this paper, we address the problem of group activity recognition (GAR) in a hierarchical way by first examining individual person’s actions, reconstructed by correlating data coming from body-worn and external positioning sensors. We then aggregate this information by considering each individual as an input of a hierarchical deep belief network (DBN). This aims to extract common temporal/spatial dynamics at the level of group activity. We evaluated the proposed approach in a laboratory environment, where the participants labeled their daily activities using an app on a mobile phone. Collected data contributed to the creation of two datasets respectively containing labeled single and group activities. The experimental results evaluated on these datasets and on a public one demonstrated the effectiveness of the proposed model with respect to a support vector machine (SVM) baseline.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11367/97657
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 5
  • ???jsp.display-item.citation.isi??? 4
social impact